Artificial neural network-based standalone tunable RF sensor system

Abstract

An artificial neural network (ANN) based tunable automated standalone RF sensor system is proposed to realize an improved sensing module involving a quite accurate solution of the non-linear inverse characterization problem. The presented tunable sensor system is quite novel as it alleviates the need for any active tuning circuitry. Moreover, the proposed unified design topology facilitates a relatively higher tuning range (1900MHz) than that of the earlier reported (580MHz) capacitor-based tunable complementary split-ring resonator (CSRR). The higher tuning range of structures resulted from the improved design configuration comprising a modified CSRR design coupled with a modified microstrip line. The obtained dielectric sensitivity is ∼8.8%. The numerically generated S-parameters of various dielectric samples are used here as a training dataset for the ANN, which is trained using the Levenberg-Marquardt backpropagation algorithm in combination with the Bayesian regularization. Finally, several standard test samples at different unloaded tuned frequencies are measured to record the corresponding resonant frequency and magnitude of the S-parameter in order to process them using the proposed ANN-based sensor system. It is found that the developed ANN-based sensor system provides a reasonably accurate value of the extracted complex permittivity over the frequency range under consideration, which basically removes the need for designing multiple resonant structures unlikely to the conventional resonant sensors.